Identifying best feature subset for cardiac arrhythmia classification

This paper presents a model for diagnosis of cardiac arrhythmias. The model uses k-Nearest Neighbors (KNN) and support vector machines (SVM) as the classification algorithms and improved F-score and sequential forward search (IFSFS) as the feature selection method. Complete feature selection process presented by this paper comprises two parts i.e. filter part and the wrapper part. Improved F-score is the criterion used in the filter part of the model, and in the wrapper part SFS is used. The setup uses KNN and SVM alternatively in the wrapper part to obtain the best feature subset. 20 fold cross validation was performed on Arrhythmia dataset obtained by UCI (University of California Irvine) machine learning repository. Experiments show that presented model achieves average accuracy of 73.8% in case of KNN, and 68.8% in case of SVM; which makes the model outperform the accuracies achieved by previous methodologies.